Roel Veerkamp & Claudia Kamphuis · 2019. 1. 15. · MY MY Variance in deviations Lag-1...
Transcript of Roel Veerkamp & Claudia Kamphuis · 2019. 1. 15. · MY MY Variance in deviations Lag-1...
Big Data in de melkveehouderij
Roel Veerkamp & Claudia Kamphuis
What is Big Data ?
1.79 billion 317 millionmonthly active users
Big Data and the 6Vs
Capability to acquire, understand, and interpret data real-time
3
Forms of (un)structured data (spreadsheets, text, tweets, video, drone images)
Reliability and quality of data
Data whose meaning is constantly changing
Expectations are huge if analysis of Big Data delivers insights and information
Volume
Velocity
Variety
Veracity
Variability
Value
Big data ........
Challenge application Big Data
5
Domain knowledge
Data analytics
ICT skills
Big Data & Wageningen Livestock Research
6
Management tools
Sensor technologies
Food chain
Three examples of Big Data projects
Predict dairy cow’s longevity
Resilience and efficiency of animal and farms
More specific manure application (norms): Bemestwijs
Big Data & Wageningen Livestock Research
Example project 1
Predict dairy cow’s longevity
8
Predicting cows longevity
Longevity of a cow important: economics, management and society.
Predict expected longevity of an animal?
9
Predicting cows longevity
DNA van 6847 calves on 463 farms used to predict phenotype: breeding value for 50 traits
72 additional phenotypic records; Pedigree, dam, own birth and calving records, test milk days, movement (transport), inseminations, viability & vitality of calves, survival status at various points, farm...
Statistical methods: Machine learning
10
Predicting cow longevity
Breeding value of lifespan, and one or more from;
● fertility
● udder health’, conformation
● feet and legs (foot angle)
● body conformation (size)
Important phenotypic traits are;
● Season of birth and calving
● Fertility traits (insemination#, NR status)
● Age at first calving
● Milk production (kg)
● Udder health (cell score)
11
Predicting cows longevity
top 50% heifer calves are selected:
12
Big Data & Wageningen Livestock Research
Example project 2 Gentore:
Efficiency and resilience at cow and farm level
Claudia Kamphuis, Wijbrand Ouweltjes, Yvette de Haas
13
WP3 On-farm phenotyping
14
Near or far-off market technologies
Big Data across farms
At-market technologies
Resilience
Resilience through the theory of critical transitions
15
Scheffer et al., 2012
Stable
state 1
Stable
state 2
Perturbation
Stable
state 1
Stable
state 2
Perturbation
Heat stress: resilant farm systems
16
July 2014 June 2015
How to measure resilience
using existing data
Resilient
Not resilient
17
Disturbance
Disturbance
MY
MY
Variance in deviations
Lag-1 autocorrelation of
deviations
Skewness of deviations
Big Data & Wageningen Livestock Research
Example project 3
Field specific phosphate application norms
Erwin Mollenhorst, Claudia Kamphuis, Gerard Migchels
18
Hackatons
19
(Be)MestWijs won de aanmoedigingsprijs
voor meest marktrijpe resultaat. vlnr Job de
Pater (NMI), Reinier Wieringa (EZ-Dictu),
Erwin Mollenhorst (WUR), Justin Steenhuis
(VAA ICT), Herbert Meuleman (CRV),
Claudia Kamphuis en Gerard Migchels
(beide WUR). Niet op de foto: Roel
Veerman (Akkerweb)
2017
2018
Winnaars van de #Bodemhack
op de Marke, samen werken
aan een data- en IT-instrumentarium
om ecosysteemdiensten en resultaten
zichtbaar en meetbaar te maken
(Be)MestWijs stapsgewijs naar kunstmestvrij
Bedrijf -
KringloopwijzerPercelen -
Akkerweb
Percelen –
Precisie bemesting
Nu Korte Termijn Middellange & Lange Termijn
First trials:
Currently:
● Fixed phosphate application norms for crops / grassland
● 3 classes, based on P status of field
● For crops: 50/60/75 kg P2O5
Can we predict future maize yields (= P) based on farm data and open source weather data?
21
Dataset from “KTC De Marke”
162 records of maize yields
24 different fields
Years 1996 – 2014
On average 7 times maize
Information on:
● N and P input and output
● Irrigation, P status of field
● Weather data (own weather station and open source)
22
Variable importance (average 5 years)
4 most important variables
● Crop in previous year (grass/maize) (0.99)
● Phosphate status field (0.55)
● Maximum temperature in July (0.36)
● Average Pyield maize on same field in past 7 years (0.32)
Machine learning is marginally better in predicting P yield than a generic norm (similar RMSE)
23
Summary
More and more big data will come available (6xVs)
Technology will allow us to use data in management, better use sensors and connection in food production chain
Replace some of the classic ways of working
Technology is not the silver bullet!
24
Take home
Success Big Data is not about
technical tools, but
connecting the tools with
people and domain expertise
25